Add Graph difference

[nahumsa]

Related to #440

Adds the difference between two graphs for PyDiGraph and PyGraph.

[coveralls]

Pull Request Test Coverage Report for Build 2779975375

• 62 of 62 (100.0%) changed or added relevant lines in 2 files are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage decreased (-0.001%) to 97.131%

---

Totals
Change from base Build 2778191284:	-0.001%
Covered Lines:	12425
Relevant Lines:	12792

---

💛 - Coveralls

[mtreinish]

Overall this LGTM I think this is pretty close, I just have a small question inline about the node weights. The only thing i think that's really missing is in the docstrings for the new function is an explanation of what the difference is, an example might go along way to show what the function does and how it works.

[mtreinish]

What happens if the node weights are different between first and second? Do we care at all that if first[0] != second[0] we only preserve the payload for first[0]?

[nahumsa]

Do you think we should have a callback to handle weights from both graphs?

[mtreinish]

I think the big thing missing here is the constraints on the input types (they have to have identical node indices). This probably should mention that and also maybe have an example and/or an explanation of how the difference is computed.

[nahumsa]

Sure! I'll add examples and explain how we compute the difference in the docs.

[georgios-ts]

Users have only indirect control over node indices since we assign them internally. In union we distinguish two nodes/edges based on their data payloads and it's probably better to do the same in this PR too.
In more detail I'm thinking that we should:

1. Verify that we can match every node of the first graph into a unique node in the second graph with the same data payload.
2. Remove an edge e = (u, v) from the first graph if the second graph has an edge with endpoints e' = (m[u], m[v]), where m is the matching found in step (1.), and the edge data payloads agree.